Sverklo

v0.19.0Developer Toolsstable

Local-first code intelligence MCP server with hybrid BM25 + ONNX vector search, symbol-level impact analysis, diff-aware PR review with risk scoring, and persistent memory tied to git state.

ai-agentsbm25claude-codecode-intelligencecode-review
Share:
60
Stars
0
Downloads
0
Weekly
0/5

What is Sverklo?

Sverklo is a Model Context Protocol (MCP) server that allows AI assistants like Claude, Cursor, and VS Code to local-first code intelligence mcp server with hybrid bm25 + onnx vector search, symbol-level impact analysis, diff-aware pr review with risk scoring, and persistent memory tied to git state.

Local-first code intelligence MCP server with hybrid BM25 + ONNX vector search, symbol-level impact analysis, diff-aware PR review with risk scoring, and persistent memory tied to git state.

This server falls under the Developer Tools category on MCPgee, the world's largest MCP server directory with 33,000+ servers.

Features

  • Local-first code intelligence MCP server with hybrid BM25 +

Use Cases

Search code using hybrid BM25 and vector search.
Analyze code impact and perform PR review with risk scoring.
Maintain persistent memory tied to git state.
sverklo

Maintainer

LicenseMIT
Languagetypescript
Versionv0.19.0
UpdatedMay 21, 2026
Statushealthy
Maintenanceactive

Works with

ClaudeOpenAIwindowsmacoslinux

Installation

NPM

npx -y sverklo

Manual Installation

npx -y sverklo

Configuration

Configuration Details

Config File

claude_desktop_config.json

Performance

Response Metrics

Response Time< 200ms
ThroughputMedium

Resource Usage

Memory UsageLow
CPU UsageLow

How to Set Up and Use Sverklo

Sverklo is a local-first MCP server that gives AI coding agents (Claude Code, Cursor, Windsurf, Codex CLI) persistent memory of your repository's structure and history. It combines hybrid BM25 and ONNX vector search using a locally cached all-MiniLM-L6-v2 model, a symbol-level impact graph for blast-radius analysis, diff-aware PR review with risk scoring, and bi-temporal memory tied to git SHAs — all without sending any code to an external service. With 37 MCP tools and sub-second incremental refresh, it replaces ad-hoc grep with ranked, relationship-aware code retrieval that scales to large monorepos.

Prerequisites

  • Node.js 18.x or higher (for npx/npm install)
  • Git repository initialized in your project directory
  • An MCP-compatible coding agent: Claude Code, Cursor, Windsurf, VS Code with MCP support, or any MCP-speaking client
  • ~86 MB disk space for the ONNX embedding model (downloaded once to ~/.sverklo/models/ on first run)
  • No API keys required — fully offline after first model download
1

Preview what Sverklo finds in your repo

Run the prove command in dry-run mode to see a sample of real repo-memory output from your own codebase before changing any config. This downloads the ONNX model on first run and caches it under ~/.sverklo.

cd your-project
npm exec --yes --package=sverklo@latest -- sverklo prove --no-write --guided --markdown
2

Preview the setup changes

Run init in dry-run mode to see exactly which files (MCP config, CLAUDE.md / AGENTS.md) would be created or modified before committing.

npx -y sverklo init --dry-run
3

Install globally and initialize

Install sverklo globally, then run init. It auto-detects your installed coding agent, writes the correct MCP config, appends instructions to AGENTS.md or CLAUDE.md, and runs sverklo doctor to verify the handshake.

npm install -g sverklo
sverklo init
4

Configure your MCP client manually (alternative)

If you prefer manual setup over sverklo init, add this entry to your MCP client config. No environment variables are required.

{
  "mcpServers": {
    "sverklo": {
      "command": "npx",
      "args": ["-y", "sverklo"]
    }
  }
}
5

Verify the setup

Ask your coding agent to list available sverklo tools, or run the doctor command to confirm the MCP server is connected and indexed.

sverklo doctor

Sverklo Examples

Client configuration

Minimal MCP config for any MCP-compatible client. No API keys or environment variables needed — Sverklo is fully local.

{
  "mcpServers": {
    "sverklo": {
      "command": "npx",
      "args": ["-y", "sverklo"]
    }
  }
}

Prompts to try

Example queries for your coding agent that leverage Sverklo's symbol graph, vector search, and memory tools.

- "Search the codebase for authentication flow handling."
- "Show me all callers of the UserService.validate method and their files."
- "What is the blast radius if I change the BillingAccount.charge function?"
- "Review the current git diff and rank the changed files by risk."
- "What architectural decisions were recorded about the auth module?"
- "Which files are most load-bearing in this codebase based on the dependency graph?"

Troubleshooting Sverklo

First run is slow or the prove command hangs for several minutes

The all-MiniLM-L6-v2 ONNX model (~86 MB) is downloaded from HuggingFace on first use. This is a one-time download cached to ~/.sverklo/models/. Subsequent runs are fully offline and fast.

sverklo doctor reports 'MCP handshake failed'

Ensure your coding agent has been restarted after the MCP config was written. In Claude Code, run /mcp to check connected servers. For Cursor, restart the IDE after editing .cursor/mcp.json.

Search returns irrelevant results or misses known symbols

Run 'sverklo index' to force a full re-index of the repository. The incremental index updates on file changes, but a manual re-index is needed after large rebases or when adding new file types. Check 'sverklo doctor' to see indexing status.

Frequently Asked Questions about Sverklo

What is Sverklo?

Sverklo is a Model Context Protocol (MCP) server that local-first code intelligence mcp server with hybrid bm25 + onnx vector search, symbol-level impact analysis, diff-aware pr review with risk scoring, and persistent memory tied to git state. It connects AI assistants to external tools and data sources through a standardized interface.

How do I install Sverklo?

Install via npm with the command: npx -y sverklo. Then add the server configuration to your AI client's JSON config file (e.g., claude_desktop_config.json or .cursor/mcp.json).

Which AI clients work with Sverklo?

Sverklo works with all major MCP-compatible AI clients including Claude Desktop, Claude Code, Cursor, VS Code (GitHub Copilot), Windsurf, and Cline.

Is Sverklo free to use?

Yes, Sverklo is open source and available under the MIT license. You can use it freely in both personal and commercial projects.

Browse More Developer Tools MCP Servers

Explore all developer tools servers available in the MCPgee directory. Each server includes setup guides for Claude, Cursor, and VS Code.

Quick Config Preview

{ "mcpServers": { "sverklo": { "command": "npx", "args": ["-y", "sverklo"] } } }

Add this to your claude_desktop_config.json or .cursor/mcp.json

Read the full setup guide →

Ready to use Sverklo?

Browse our complete directory of 33,000+ MCP servers, read setup guides for your editor, and start building with the Model Context Protocol.

33,000+ ServersFree & Open SourceStep-by-Step Guides